-
Notifications
You must be signed in to change notification settings - Fork 0
All issues
Issue creation is restricted in this repository
Issues
is:issue state:open
is:issue state:open
Search results
GLM-5.3 TP2: enable and validate software-unlocked PCIe BAR1 P2P
enhancementNew feature or requestNew feature or requestStatus: Open.#39 In ro99/strata;- Status: Open.#38 In ro99/strata;
- Status: Open.#37 In ro99/strata;
- Status: Open.#36 In ro99/strata;
- Status: Open.#34 In ro99/strata;
- Status: Open.#33 In ro99/strata;
Unmeasured: whether tensor-parallel decode targets the bottleneck for any non-DeepSeek model
questionFurther information is requestedFurther information is requestedStatus: Open.#32 In ro99/strata;No shared execution abstraction: six standalone runtimes make every cross-cutting feature cost N implementations
enhancementNew feature or requestNew feature or requestStatus: Open.#31 In ro99/strata;Tensor-parallel decode is DeepSeek-only; no model-agnostic TP layer exists
enhancementNew feature or requestNew feature or requestStatus: Open.#30 In ro99/strata;Rank-local benchmark scripts do not pin CUDA_DEVICE_ORDER, so --devices selects different GPUs per shell
bugSomething isn't workingSomething isn't workingStatus: Open.#29 In ro99/strata;Rank-local TP2 is structurally two-rank and requires two NUMA nodes, excluding 4-GPU and single-socket machines
enhancementNew feature or requestNew feature or requestStatus: Open.#28 In ro99/strata;No capability-aware GPU selection: mixed-hardware machines require the user to know CUDA ordinals
enhancementNew feature or requestNew feature or requestStatus: Open.#27 In ro99/strata;